Jan Lindström <jplindst@mariadb.org> writes:
Yes, I did run using valgrind, while it reported some uninitialized variables, not really any memory corruptions. Attached, the log from node0. Similarly,
Well, uninitialised variables are exactly what I would expect could lead to things looking like you have shown so far. So if you have Valgrind errors, you should definitely start by fixing them. However, I see no messages from Valgrind in the log you attached?!? So please, repeat the exact problem you get (assertion during rollback) while running with Valgrind. Provide the exact command you used to get the error in this particular run, and the correct log that shows the Valgrind errors about uninitialised variables, and I can give more suggestions for how to fix.
debug=+d did not help. Where I should add some output to narrow down the place where the corruption happens.
What do you mean "did not help"? Did it not work to produce any output? Maybe a different syntax is needed (I usually enable it using ./mtr --debug, but I did test using the above syntax directly). Once you get dbug output enabled, you get a file containing the output from every DBUG_ENTER() and DBUG_PRINT() statement in the code (likely many megabytes worth). This will help show what the code was doing before it crashed, so we can backtrack in the source and try to find the root of the issue. So again, repeat the exact problem you get with dbug enabled. Provide the exact command you use to repeat the problem in this exact run. And provide the resolved stack trace in GDB of the exact same failure run. And upload the DBUG output somewhere (it will be too big to mail). Then I can take a look and see if I can come up with more suggestions. - Kristian.